Nvidia adds Cosmos simulation checkpoints and inference containers

Nvidia says new Cosmos simulation recipes and released model checkpoints let developers generate synchronized camera video and LiDAR observations for autonomous vehicle testing, with additional models for robot and surgical simulation. Its new Cosmos NIM package serves supported models through application APIs using optimized containers. The driving models can vary weather and lighting while preserving road layouts and vehicle trajectories, or generate observations incrementally as driving inputs change. Robot models support visual variation and interactive two-arm simulation; surgical models generate video from robot actions for policy evaluation before physical testing. The simulation checkpoints support PyTorch and vLLM-Omni and were tested on Nvidia Hopper and Blackwell GPUs.